Dual 2.2Ghz P4 and Hyper Threading for SETI (UPDATED)

[barbary]

Some of you have contacted me about my new workstation.

Well it finally arrived today. Nice people at target express here in england have been sitting on it in there warehouse since saturday.

And yes it does have hyper threading options in the bios.

So I've whipped up some fast stats for you to debate.

I'll be doing some specific seti testinf later as requested but I have to do some work first. Plan is to find a "typical" WU. Then process it over and over (Never putting it back) so I can test different options with a common reference.

These results are now added in further down the thread

For your enjoyment here is some quicky clibench MkIII 0.7.7 benchmarks.

No Hyper
2 Threads
Dhrystone 2.1 .............. 3848
Whetstone .................... 1126
Eight queens problem .... 8178
Matrix operations .......... 157727
Number crunch .............. 275983
Floating point ................ 17065
Memory throughput ...... 405120


Hyper Thread
2 Thread

Dhrystone 2.1 ................. 3424
Whetstone ...................... 987
Eight queens problem ..... 7950
Matrix operations ............ 150237
Number crunch ................ 278436
Floating point .................. 13517
Memory throughput ........ 405120

Hyper
4 Threads

Dhrystone 2.1 ............... 3413
Whetstone .................... 1509
Eight queens problem ... 7949
Matrix operations .......... 222193
Number crunch .............. 294855
Floating point ................ 14379
Memory throughput ...... 324096

Hyper
6 threads

Dhrystone 2.1 ................ 3588
Whetstone ..................... 1472
Eight queens problem .... 8320
Matrix operations ........... 225964
Number crunch ............... 292956
Floating point ................. 15178
Memory throughput ....... 313641

Note although there is a small drop off in some test. Whetstone and Matrix operations are 30 to 40 per cent faster.
These results now finished and added in reply belowThese resultsText

 

[barbary]

In case anybody else has a Dell 530 get the updated bios with Hyper Thread Options here

http://support.dell.com/us/en/filel...&os=WW1&searchtype=filter&devid=258&type=BIOS

 

[Smoke]

Benchmark File

 

[barbary]

Now that is a good idea.

I'll have to start again.

 

[Assimilator1]

I look forwards to the WU times on that

 

[Sukhoi]

I'm actually not surprised the scores on those hyper-threaded bechmarks are lower. From what I could tell in Anand's article, hyper threading can actually slow the system down a little if a lot of the same stuff is being done.

That's why I'm interested to see how SETI and RC5, or SETI and ECCp-109 work together. SETI is floating point, while RC5 and I believe ECCp-109 are integer based. So theoretically they should run fairly well together since they use different parts of the processor.

Nice machine though!

 

[barbary]

Well it's taken a while but results are in. They also very interesting even if I do say so myself.

All of these tests used the standard benchmark WU. (Find it in a link from Smokeball above) Although I left the machine alone there are a few things running on it. Nothing that would make a difference of more than a minute on result times. All times have been rounded up to the nearest minute.

2 Seti Cli no Hyper Threading
Elapsed time 3:42, Results 2, Time per Unit Per CPU 3:42

4 Seit Cli no Hyper Threading
Elapsed time 7:22, Results 4, Time per Unit Per CPU 3:41

4 Seti Cli with Hyper Threading
Elapsed time 5:47, Results 4, Time per Unit Per CPU 2:54

Just for Comparrison my 1700Mhz Xeon has a, Time per Unit Per CPU 4:37 without Hyper Threading.

The normal running mode time I found disapointing the extra 500Mhz per CPU had taken rouughly 1 hour off a WU time. With 512k full speed L2 cache I would have expected or at least hopped for a little more. This equates to about 1.7 WU's extra a day per CPU where as a 400Mhz PII is able to produce about 1.9 WU a day. Which still shows that per clock the older processors still do more.

The first set of results Is expected from a normal 2 CPU machine. With 4 Cli the time is doubled with each processor having twice as much to do. Any swaping overheading being very un-noticable at this sort of CPU speed.

With hyper threading switched on it seems to perform indivual tasks slightly worse. (Check back at the cli bench marks above) But the overal effect when the CPU is fully loaded is an increase in performance. As the Jackson method keeps the CPU more stacked with instructions to do.

 

[ElFenix]

jebus under 3 hours!


i think thats a record, you should turn it in

 

[BadThad]

Thanks for the info! Very interesting.

 

[Assimilator1]

Time per Unit Per CPU 2:54

Nice !
Re the PII 400 vs P4 efficiency issue ,goes to show what the longer pipeline does

 

[Sukhoi]

Very nice times!

Could you run a SETI WU with RC5 going at the same time and see how much each slow down?

 

[IndyJaws]

Time per Unit Per CPU 2:54 :Q

 

[Engineer]

/me faints! :Q

I've got to get something faster...these Dual 1.4GHz TBirds are the pits now...hehe!

 

[Baldy18]

If I'm readin that right that is ~16.7 WU/day from that one dually box? WOW!

 

[DarkMajiq]

:Q

 

[barbary]

It's had 2 full days of normal running. I'm getting about 18/19 WU's a day from the box.

Elaspsed time is running around 5.10. Of course each processor does 2 WU's in that time giving a WU time of 2:35.

 

[PieDerro]

FAAAAAAAAAAAAAR OUT!!!!!!!

Which *ahem* affordable P4 boards are out there that can turn on Hyper Threading? Even for single CPU!? I ask this cos I can get my hands on a P4 1.8GHz Northwood at the good price they are right now, crank it up to 2.6GHz or so on a DDR motherboard, turn on Hyper Threading and watch those WUs flyyyyy!!!!

droooooooool... hehehe...

 

[barbary]

When I first started seti on Mon May 17 16:22:33 1999 UTC I had a dual 400 PII on my desk and it used to pump out a whopping 4/5 a day.

Less than 3 years later and the machine on my desk turns out 18/19.

Goes to show how far desktop processors have come.

 

[lane42]

Good times barbary. 2:54 on a .417 wu, My 1900+ does lots of 2:50 times but not as an ave. And ill wish you luck on getting your times on ars, ive been waiting for months :frown:

 

[Assimilator1]

PieDerro
I thought the Northwoods only started at 2.0 GHz?

 

[PieDerro]

Assimilator1, actually, this place here in Australia sells 1.6GHz and 1.8GHz Northwoods!! Not engineering samples either. My guess is that they are parts being manufactured for the Dells etc of this world. They go for the same price as their Willamette counterparts.

 

[barbary]

Here's the responce from ars. Seem to have problems explaining that you have two threads on one processor. There realy isn't an acurate way to display this on his site yet.

I was thinking of posting 2 submissions: 2 instances on 2 CPUs and 4
instances on 2 CPUs. I have to report the cpu time given by the S@H client,
so for instance 1 it will be 13270.062500 seconds (3:41) and for instance 2
it will be 20826.218750 seconds (5:47).

Roelof

 

[Poof]

Barbary - consider yourself forcing Roelof and Max to blaze new trails! LOL!

And also think about this... A year ago today (Feb. 1, 2001), we were all forced to migrate to 3.03. This was due to:

1.) Berkeley network overload
2.) Berkeley campus bandwidth restrictions implemented
3.) ALOT of fast WU returns, ie., many under 4 hours.

And whaddya know? It looks like it's deja vu all over again.

A year ago, we *knew* that cop-outs like doubling the WU times, weren't gonna fix the "problems" of faster and faster CPUs. And so... here we are again at the same point that we were last year. Unable to reliably connect to SETI to upload results. :disgust:

 

[Assimilator1]

It did work..............for about a year
Btw it didnt double the WU time for the most part but added about 50% ,which still sucks though